Low Dimensional Reproduction Strategy for Real-Coded Evolutionary Algorithms

The strategy of low dimensional reproduction (LDR) is proposed for real-coded evolutionary algorithms (REAs) in this paper. It preserves some (randomly chosen) components of the local best vector (elite individual) in the reproduction process and let the traditional reproduction operators act on the rest components. Thus it could help the search points escape from the hyperplane where the parents individuals lies, as well as keep them from getting too much decentralized and search mainly along a series of orthogonal directions (coordinate). The LDR strategy provides a universal idea to improve the performance of REAs. Four REAs are taken as examples to show the effect of the strategy. Numerical results show that the proposed strategy can accelerate the convergence speed of the applied algorithms considerably. In addition, the strategy is computational saving, easy to implement, and easy to control.